New AI Trick Makes Your Photos Smaller Without Losing Quality
This could double your phone's photo storage and speed up image sharing.
Neural image compression typically uses uniform scalar quantization because vector quantization is highly complex. A new paper proposes OLVQ, a learning method that designs rate-distortion optimal lattice vector quantization codebooks based on the sample statistics of latent features to be compressed.
By better fitting LVQ structures to any given latent sample distribution, OLVQ significantly improves the rate-distortion performance of existing quantization schemes in neural image compression while retaining the amenability of uniform scalar quantization. Code is available at the provided GitHub link.
- OLVQ is a new AI-based compression method that makes image files smaller without losing quality.
- It adapts to the specific patterns in each image, unlike traditional methods that use fixed rules.
- In tests, it outperformed existing techniques, potentially doubling storage capacity for photos.
Why It Matters
Smaller image files mean more photos on your phone and faster loading websites, saving data and storage.